Seeing is Deceiving: The Rise of AI Synthesized Fake Media

### The Technological Evolution of Deepfakes At 00:00, Professor Siwei Lyu from SUNY Albany presents on the rapid technological evolution and threat landscape of artificial intelligence-synthesized media (deepfakes) at the AI Village. Professor Lyu reviews how generative models create hyper-realistic synthetic video.

### Generative Architectures: Encoders, Decoders, and GANs At 07:29, Professor Lyu details the core machine learning architectures powering face swapping. He explains shared encoder and individual decoder pipelines, showing how models compress facial expressions and synthesize them onto target identities using Generative Adversarial Networks (GANs).

### Identifying Biological and Physical Artifacts At 15:02, the presentation explores forensic telltales found in early synthetic media. Professor Lyu details how discrepancies in spontaneous physiological signals—such as irregular eye blinking rates, missing corneal reflections, and unnatural head-pose angles—allow forensic analysts to detect altered media.

### The Democratization of Disinformation At 22:34, the speaker highlights the societal impact of accessible deepfake generation software. He expresses concern over how publicly available social media photos can be scraped and weaponized to generate non-consensual imagery, financial extortion schemes, and political disinformation.

### The Cat-and-Mouse Game of Synthetic Media Forensics Concluding between 30:12 and 38:11, Professor Lyu evaluates the ongoing competition between deepfake generators and forensic detection algorithms. He emphasizes the need for cryptographic provenance standards, watermarking, and cross-disciplinary legal policies to safeguard digital media trust.

(AI summary — Gemini Flash)